Predibase

Profile (continued)

PRODUCT EVOLUTION: Predibase launched publicly as a 'low-code, declarative machine learning platform' aimed at letting data scientists and engineers build and deploy ML models (including but not limited to LLMs) with minimal boilerplate code, an extension of Ludwig's declarative, YAML-driven philosophy into a managed commercial product. As the generative AI wave accelerated in 2023, Predibase pivoted its narrative and product emphasis toward LLM fine-tuning specifically, introducing 'LudwigGPT' style LLM fine-tuning support, a 'Data Science Copilot' for performance recommendations, and (by 2023-2025) a proprietary post-training stack plus an optimized inference engine claimed to deliver roughly 2x performance gains and up to 80% lower infrastructure costs versus naive serving. A cornerstone open-source project, LoRA eXchange (LoRAX), lets Predibase (and any user) run thousands of fine-tuned LoRA adapters on a single GPU by dynamically swapping lightweight adapter weights against a shared base model — as of the 2025 acquisition LoRAX had roughly 3,800+ GitHub stars. 'Turbo LoRA' combined LoRA fine-tuning with speculative decoding techniques to further speed inference, in some cases reportedly tripling throughput.

FUNDING HISTORY: Predibase raised a total of approximately $28.45 million in disclosed venture funding across what press coverage refers to somewhat inconsistently as one or two 'Series A' tranches (Greylock's own portfolio page instead labels the earliest round 'Seed (First Partnered),' i.e., Greylock led/co-led the company's first institutional financing). An initial round of $16.25 million (reported by TechCrunch as a 'Series A' completed in the prior year, i.e. circa 2022, though the exact announcement date could not be independently confirmed from available sources) was followed by an additional $12.2 million raised and announced around May 31–June 1, 2023, led by Felicis Ventures, bringing cumulative funding to $28.45 million (16.25 + 12.2). This total ($28.45M, rounded to '~$28 million' by NBC/CNBC coverage of the Rubrik deal) is corroborated by two independent sources (TechCrunch's math and NBC's acquisition coverage). Confirmed investors include Greylock Partners (earliest/lead backer per Greylock's own site), Felicis Ventures (lead on the 2023 extension, via partner/board observer Niki Pezeshki), and individual angel investor Anthony Goldbloom (co-founder and former CEO of Kaggle), per Crunchbase. Because Crunchbase's detailed funding-round table was paywalled/obfuscated during this research, exact per-round dates, valuations, and the complete investor participant list for each individual tranche could not be fully verified beyond what is stated above; those specific gaps are represented as nulls rather than guessed.

EXIT — ACQUISITION BY RUBRIK: On June 25, 2025, cybersecurity and data-management company Rubrik, Inc. (NYSE: RBRK) announced it had agreed to acquire Predibase to accelerate 'agentic AI adoption,' pairing Rubrik's secure enterprise data infrastructure with Predibase's LLM fine-tuning and serving stack. The financial terms were not officially disclosed by either company; multiple outlets (SiliconANGLE citing CNBC, NBC/CNBC's own reporting) put the deal value at over $100 million, with some reports citing a broader range up to roughly $500 million — these figures should be treated as press estimates, not confirmed figures. Rubrik CEO Bipul Sinha said 'What the Predibase team has achieved with model training and serving infrastructure in the last few years is nothing short of remarkable... Together, Rubrik and Predibase will drive agentic AI adoption around the world.' Predibase CEO Devvret Rishi said 'Predibase removes the hardest part of that journey and accelerates production-ready AI by giving teams an easy-to-use platform to tune models to their own data.' Post-acquisition, Predibase was expected to operate as a standalone business unit within Rubrik, maintaining compatibility with its existing third-party data and cloud integrations (AWS, Azure, Google Cloud / Google Agentspace, Amazon Bedrock, Azure OpenAI). Crunchbase's operating status field for Predibase now reads 'Closed' (i.e., independently defunct as an acquired subsidiary), consistent with an 'acquired' status in this record.

MARKET POSITION / COMPETITION: Predibase operated in the crowded LLM infrastructure / MLOps and inference-serving space. Crunchbase's own 'similar companies' listing names CoreWeave, NVIDIA, and Fireworks AI as competitors; based on general market knowledge, other companies competing in adjacent fine-tuning/serving/MLOps niches around the same period included Together AI, Anyscale, Databricks (Mosaic AI / former MosaicML), Hugging Face, Modal, Baseten, and Replicate, though these broader comparisons were not individually sourced beyond general industry knowledge and should be treated as illustrative context rather than confirmed head-to-head competitors.

OVERALL: Predibase is best understood as a direct commercial descendant of the open-source Ludwig project born at Uber, taken to market by its creators (with Stanford's Chris Ré as an academic co-founder), backed first by Greylock and then by Felicis, and ultimately acquired by Rubrik in 2025 to become that company's enterprise agentic-AI/LLM fine-tuning unit.

Go-to-market

Enterprise AI/ML teams and startups seeking to fine-tune and deploy open-source LLMs and traditional ML models in production; named customers/users include Paradigm and Koble.ai, with additional reported reach into Fortune 500 organizations.

Ownership

subsidiary — parent: Rubrik, Inc.